Re: Polyphase merge is obsolete

On 21.11.22 00:57, Heikki Linnakangas wrote:
> On 19/11/2022 13:00, Peter Eisentraut wrote:
>> On 18.10.21 14:15, Heikki Linnakangas wrote:
>>> On 05/10/2021 20:24, John Naylor wrote:
>>>> I've had a chance to review and test out the v5 patches.
>>>
>>> Thanks! I fixed the stray reference to PostgreSQL 14 that Zhihong
>>> mentioned, and pushed.
>>
>> AFAICT, this thread updated the API of LogicalTapeSetCreate() in PG15,
>> but did not adequately update the function header comment.  The comment
>> still mentions the "shared" argument, which has been removed.  There is
>> a new "preallocate" argument that is not mentioned at all.  Also, it's
>> not easy to match the actual callers to the call variants described in
>> the comment.  Could someone who remembers this work perhaps look this
>> over and update the comment?
>
> Is the attached more readable?

That looks better, thanks.

> I'm not 100% sure of the "preallocate" argument. If I understand
> correctly, you should pass true if you are writing multiple tapes at the
> same time, and false otherwise. HashAgg passed true, tuplesort passes
> false. However, it's not clear to me why we couldn't just always do the
> preallocation. It seems pretty harmless even if it's not helpful. Or do
> it when there are multiple writer tapes, and not otherwise. The
> parameter was added in commit 0758964963 so presumably there was a
> reason, but at a quick glance at the thread that led to that commit, I
> couldn't see what it was.

Right, these are the kinds of questions such a comment ought to answer.

Let's see if anyone chimes in here, otherwise let's complain in the
original thread, since it had nothing to do with this one.